the effect of inflation on inflation uncertainty in the g7 ...kushal banik chowdhury and nityananda...
TRANSCRIPT
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
34
The Effect of Inflation on Inflation Uncertainty in the G7 Countries:
A Double Threshold GARCH Model
Kushal Banik Chowdhury
and Nityananda Sarkar
Indian Statistical Institute and Indian Statistical Institute
ABSTRACT
This paper studies the impact of inflation on inflation uncertainty in a modelling
framework where both the conditional mean and conditional variance of inflation are
regime specific, and the GARCH model for inflation uncertainty is extended by including a lagged inflation term in each regime. Applying this model to the G7 countries with
monthly data from 1970 till 2013, it is found that the impact of inflation on inflation
uncertainty differs over the regimes in most of the G7 countries. The findings also provide strong empirical support to the well-known Friedman-Ball hypothesis of positive
impact of inflation on inflation uncertainty, but only for the high-inflation regime.
Key words: Inflation; Inflation Uncertainty; Regime Switching Model
JEL Classifications: C22; E31
1. INTRODUCTION
The impact of inflation on inflation uncertainty is well documented in theory. The empirical
literature on this topic is also sizeable. However, the findings on the nature of the relationship
and the empirical support for the link involving these variables vary with mixed results.
Consequently, the Friedman-Ball hypothesis (Friedman, 1977; and Ball, 1992), which states
that the effect of inflation on its uncertainty is positive, has found empirical support in varying
degrees1. While differences in the sample periods and frequencies of the data sets used could
account for mixed results, more importantly, it is the methodology or modelling approach
applied that would explain the varied findings.
It is worth mentioning that in many such empirical studies, a standard auxiliary assumption
typically made is that the parameters depicting the relationship are constant over the entire
sample period. This means that the causal links are assumed to be stable over time. However,
this assumption is far from innocuous and may often not hold in practice. Furthermore, there
is considerable evidence supporting the view that the economic time series are best thought of
as being generated by processes with time dependent parameters. According to several
studies, including Fischer (1993), Stock and Watson (1996), Caglayan and Filiztekin (2003),
Arghyrou et al. (2005) and Lanne (2006), social, economic and political changes are likely to
Kushal Banik Chowdhury, Economic Research Unit, Indian Statistical Unit, 203 B. T. Road, Kolkata-700108, India, (email: [email protected]), Tel: +919051906848.
Nityananda Sarkar, Indian Statistical Institute, 203 B.T. Road, Kolkata-700108, India, (email: tanya@isical
.ac.in). 1 See, for details on the evidence of positive effect of inflation on inflation uncertainty, Grier and Perry (1998),
Fountas (2001), Fountas et al. (2002), Kontonikas (2004), Daal et al. (2005), and Fountas and Karanasos (2007),
and for negative effect, Engle (1983), Bollerslev (1986), Cosimano and Jansen (1988), Hwang (2001), and
Wilson (2006).
mailto:[email protected]:[email protected]:[email protected]
-
International Econometric Review (IER)
35
make the macroeconomic relationship with constant parameter linear models quite untenable,
especially when the length of the time series is long enough.
In fact, a recent trend in the literature on the relationship between inflation and inflation
uncertainty is that this relationship is assumed to be neither linear nor stable over time. While
there are a number of modelling procedures that incorporate these important aspects, an
important class of such models is the regime switching models. In case of inflation, regime
switching models are quite relevant and appropriate since inflation, by its very nature as a
macro variable, may change for a period of time before reverting back to its original
behaviour or switching to yet another style of behaviour, often due to intervention by
government and/ or regularity authorities. The issue of regime switching behaviour of
inflation while dealing with inflation uncertainty, was first raised by Evans and Wachtel
(1993). They concluded that adequate attention needs to be paid to the consequences of
regime switches; otherwise, existing models without any consideration to regime changes
would seriously underestimate both the degree of uncertainty of inflation and its impact.
Following their lead, some researchers including Kim (1993), Kim and Nelson (1999), Bhar
and Hamori (2004), Chang and He (2010), and Chang (2012) have studied the relationship
between inflation and its uncertainty where the regime specific behaviour have been duly
incorporated in the model. A point worth noting is that all the above-mentioned studies have
used the concept of unobserved regime and accordingly applied the well-known Markov
switching regression model2.
There is another class of regime switching models where regimes can be characterized by a
threshold variable that is observable (see, Tong and Lim, 1980; Chan and Tong, 1986;
Granger and Teräsvirta, 1993; Teräsvirta, 1998; Li and Li, 1996; and Brooks, 2001; for details
on such models). This work, which essentially studies the impact of inflation on inflation
uncertainty, takes be the past level of inflation as a threshold variable. The notion of taking
the level of inflation as the observed threshold variable has been influenced by a few recent
studies. For instance, Baillie et al. (1996) applied an autoregressive fractionally integrated
moving average (ARFIMA) model to describe the long memory process of inflation dynamics
for ten countries. They found that for six low inflation countries viz., Canada, France,
Germany, Italy, Japan, and the USA, there is no apparent relationship between mean and
variance of inflation. However, for the high inflation economics of Argentina, Brazil, Israel,
and the UK, there is strong support for the Friedman hypothesis. In a recent study, Chen et al.
(2008) have observed that the effect of inflation on inflation uncertainty is asymmetric. To be
more specific, they have found a U-shaped pattern for four countries of East Asia, based on a
nonlinear flexible regression model of Hamilton (2001), suggesting that inflation uncertainty
is more sensitive to inflation in an inflationary period than in a deflationary period. Thus, it is
relevant as well as important to examine the dynamic interaction between inflation and
inflation uncertainty in the different regimes considered, and also to find if these interactions
are different across the different regimes.
The purpose of this study is to examine the effect of inflation on inflation uncertainty in a
regime switching modelling framework where regimes are determined by the past level of
inflation. This model has two other additional features as well. The first is that not only the
conditional mean model but also the conditional variance model is regime specific. Since
(G)ARCH (Engle, 1982; and Bollerslev, 1986) or similar other models are preferred when
2 In this class of Markov switching regression models, regimes are not deterministic and can be determined by an
underlying unobservable stochastic process depending upon the probabilities assigned to the occurrence of the
different regimes.
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
36
measuring inflation uncertainty, we too have applied the GARCH model; accordingly, the
proposed model is a double-threshold GARCH (DTGARCH) model, originally proposed by
Li and Li (1996) in the context of stock returns3. The second feature for our model involves
the extension of the usual GARCH specification by explicitly incorporating a lagged inflation
term in the conditional variance specification, the coefficient of which depicts the impact of
inflation on inflation uncertainty. One distinct advantage of this model is that it allows for
different behaviours of inflation uncertainty in different regimes, including the one captured
through the lagged inflation term. We have applied this model to the time series covering the
period from January 1970 to December 2013 for all members of the G7 countries. In this
context it may be mentioned that since we have considered a long enough time period, we
have specifically considered the issue of structural break in our study. However, unlike others,
we have dealt with the twin issues of structural break and stationarity together by applying the
recently developed tests of Perron and Yabu (2009) and Kim and Perron (2009).
The format of the paper is as follows: Section 2 describes the proposed model. Section 3
discusses empirical findings. Section 4 ends the paper with some concluding observations.
2. THE PROPOSED MODEL
Fountas et al. (2000) first considered the model for inflation uncertainty to explicitly include a
lag inflation term. Like most studies, they used the GARCH (1,1)4 model as the basic model
to measure inflation uncertainty and then augmented it by including the first lag of inflation.
The conditional mean model was taken to be the usual AR model. Thus, their model,
designated as the AR(k)-GARCH(1,1)L(1)5, consists of the following specifications for the
conditional mean and conditional variance.
πt =
ϕ0
+
ϕ1
πt-1
+
ϕ2
πt-2
+
⋯ + ϕk πt-k
+
εt, εt|Ψt-1
~
N(0,
hπ,t) (2.1)
hπ,t =
ω
+
αε
2t-1
+
βht-1
+
θπt-1 (2.2)
where πt stands for the value of stationary inflation at time t, Ψt-1 ={πt-1,
πt-2,…} is the
information set at t -
1, hπ,t is the conditional variance of πt at t, and k the optimal lag order in
the conditional mean specification. All roots of the polynomial (1 -
ϕ1B
-
⋯ - ϕk B
k)
=
0 are
assumed to lie outside the unit circle for stationarity of πt, and the usual restrictions on the
parameter in hπ,t are assumed to ensure positivity. Here the coefficient θ depicts the impact of
inflation on inflation uncertainty. Obviously, a significant positive value of θ provides
empirical support to the Friedman-Ball hypothesis for a given time series on inflation. We use
this model as a benchmark model to find if the introduction of regimes in both inflation and
inflation uncertainty leads to better understanding and modelling involving these two
variables.
In describing the proposed model, which is a regime-dependent model with two regimes, we
allow the effect of inflation on its uncertainty to be different in the two regimes. As already
3 See also, Chen (1998), Brooks (2001), Chen et al. (2003), Chen and So (2006), Chen et al. (2006), and Yang
and Chang (2008) for applications of the DTGARCH model for other financial variables. 4 For our study, the conditional mean model has been taken to be the AR model on the basis of behaviour of the
autocorrelation function (ACF) and partial autocorrelation function (PACF) of the inflation series. The value of k
has been determined by the Akaike’s information criterion (AIC). As regards the values of p and q of the
GARCH(p,q) specification for conditional variance, p = q = 1 has been found to be adequate, and hence
GARCH(1,1) model has been considered. 5 ‘L(1)’ stands for the fact that the specification for hπ,t includes the first lag of inflation as a separate term.
-
International Econometric Review (IER)
37
stated in the preceding section, we implicitly assume that the threshold variable which defines
the regime that occurs at any given point in time, is known, and takes the preceding value of
inflation i.e., πt-1, to be the threshold variable. We define the two regimes according to the
value of stationary inflation at t - 1, i.e., πt-1
>
0 and πt-1
≤
0, which are being labelled as high
and low inflation regimes6, respectively.
The proposed model is a generalization of the model in Fountas et al. (2000) in that here
regimes are introduced in the specifications of both the conditional mean and conditional
variance of inflation. Obviously, the proposed model is an extension of the DTGARCH
model, where the conditional variance specification includes a lag inflation term. The
resultant model, denoted as DTGARCH(1,1)L(1), is thus specified as follows:
πt =
(ϕ
l0
+
ϕ
l1πt-1
+
ϕ
l2πt-2
+
⋯ + ϕlkπt-k)
I[πt-1
≤
0]
+
(2.3)
+ (ϕ
h0
+
ϕ
h1πt-1
+
ϕ
h2πt-2
+
⋯ + ϕhkπt-k)
(1
-
I[πt-1
≤
0])
+
εt, εt|Ψt-1
~
N(0,
hπ,t)
hπ,t =
(ω
l +
α
l ε2t-1
+
β
l ht-1
+
θ
l πt-1)
I[πt-1
≤
0]
+
(2.4)
+ (ω
h +
α
h ε
2t-1
+
β
h ht-1
+
θ
h πt-1)
(1
-
I[πt-1
≤
0])
where I[.] is the indicator function which takes the value 1 if πt-1 ≤
0 and 0 otherwise, and
Ψt-1 =
{πt-1,
πt-2,…} is the information set at t
-
1. In this model coefficient vector
ξl =
(ϕ
l0,
ϕ
l1,...,
ϕ
lk,
ω
l , αl , β
l , θl )' comprises the coefficients in both conditional mean and
conditional variance specifications for the low inflation regime and similarly for the high
inflation regime ξh =
(ϕ
h0,
ϕ
h1,...,
ϕ
hk,
ω
h , αh , β
h , θh )'. Apart from the usual GARCH coefficients, θ
l
and θh denote the effects of inflation on inflation uncertainty in the low and high inflation
regimes, respectively.
The parameters of each of the benchmark and the proposed models i.e., DTGARCH(1,1)L(1)
and AR(k)-GARCH(1,1)L(1) models have been estimated by the maximum likelihood (ML)
method of estimation under the assumption of normality7 for the conditional distribution of εt.
For the iterative optimization procedure involved in the estimation process, the well-known
Berndt-Hall-Hall-Hausman (BHHH; Berndt et al., 1974) algorithm has been used. The
necessary computations have been done in GAUSS.
3. EMPIRICAL ANALYSIS
Like most of the empirical studies on the relationship between inflation and inflation
uncertainty (see, for instance, Grier and Perry, 1998; Fountas et al., 2002; Fountas et al.,
2004; Bredin and Fountas, 2005), we have used monthly data for this study. We have taken
monthly time series of consumer price index (CPI) as a measure of the monthly price level for
each of the G7 countries viz., Canada, France, Germany, Italy, Japan, the UK, and the USA.
These time series on CPI have been downloaded from the official website of Federal Reserve
Bank of St. Louis (http://research.stlouisfed.org/). Since the available time series on CPI is
not adjusted for seasonality, each series has been adjusted for seasonality by applying the
X12-ARIMA filtering method. The time period considered is from January 1970 to June
6 The reasons for this labelling are stated in the next section (p. 41). 7 It may be noted that like many financial time series, inflation has been found to be non-normal. However,
assumption of normal distribution for estimation is very common since the resulting estimates often converge
unlike some non-normal error distributions. In this particular literature on modelling inflation, the assumption of
normal distribution for the error is most common, and hence we have also made this assumption of normal
distribution for our analysis.
http://research.stlouisfed.org/
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
38
2013. The total number of observations is 522. The time series on inflation, π t, has been
obtained as the differences of the logarithm values of the CPI in percentage i.e.,
π t =log(CPIt/CPIt-1)*100
The time series of inflation for all the G7 countries are plotted in Figures 3.1. It is visually
evident that all the inflation series follow a trend - although segmented in nature, especially in
case of Canada, France, Italy, the UK, and the USA. Further, it appears from the plots that
there may be at least one structural break in each of the seven time series on inflation.
Figure 3.1 Time series plots on inflation of the G7 countries
-1
0
1
2
3
1970 1975 1980 1985 1990 1995 2000 2005 2010
Canada
-0.5
0.0
0.5
1.0
1.5
2.0
1970 1975 1980 1985 1990 1995 2000 2005 2010
France
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
1970 1975 1980 1985 1990 1995 2000 2005 2010
Germany
-0.4
0.0
0.4
0.8
1.2
1.6
2.0
2.4
2.8
1970 1975 1980 1985 1990 1995 2000 2005 2010
Italy
-1
0
1
2
3
4
1970 1975 1980 1985 1990 1995 2000 2005 2010
Japan
-1
0
1
2
3
4
5
1970 1975 1980 1985 1990 1995 2000 2005 2010
The UK
-2
-1
0
1
2
1970 1975 1980 1985 1990 1995 2000 2005 2010
The USA
We present the summary statistics on inflation, covering the sample period, for all the seven
countries in Table 3.1 below.
Country Canada France Germany Italy Japan The UK The USA
Mean 0.346 0.372 0.233 0.555 0.218 0.472 0.349
Std. dev. 0.379 0.359 0.251 0.508 0.498 0.513 0.331
Skewness 0.613 0.849 1.172 1.572 2.620 2.514 0.329
Kurtosis 4.765 3.245 7.066 5.594 14.544 14.298 6.650
Jarque-Bera 100.33* 63.92* 478.22* 360.62* 3489.17* 3319.76* 298.62*
Table 3.1 Descriptive statistics on inflation.
Notes: * indicates significance at 1% level of significance.
From this table it is evident that among the seven countries, Italy exhibits the highest average
inflation of 0.555 per cent with the standard deviation of 0.508 per cent. The three countries
viz., Italy, Japan and the UK have larger standard deviations than the remaining four
countries. The skewness and kurtosis values indicate that all the seven inflation series are
positively skewed and have leptokurtic distributions. Worthwhile to highlight are the high
values of skewness for Japan and the UK at 2.620 and 2.514, respectively. As for the kurtosis,
Japan again has the highest value of 14.544, followed by the UK with a value of 14.298. The
kurtosis values for Germany and Italy are also moderately high. Therefore, as expected,
results of the Jarque-Bera normality test clearly indicate that the null hypothesis of normality
is rejected for all the inflation series, in particular, with very high test statistic value for
countries like Japan and the UK.
-
International Econometric Review (IER)
39
We now discuss the stationarity/non-stationarity status of the inflation series. Since structural
change in a time series affects inference on unit roots of the series, we examined these two
issues together following the recently-developed procedures of Perron and Yabu (2009) and
Kim and Perron (2009).
3.1. Stationarity and Structural Break
Since 1960s most of the industrialized countries including the G7 have experienced long-term
swings in levels of inflation. Inflation had progressively risen in the 1960s and 1970s before it
declined in the 1980s. Inflation further declined in the early to mid-1990s and since then
remained low and stable (see, for details, Blinder, 1982; DeLong, 1997; Clarida et al., 2000;
Orphanides, 2003; Meltzer, 2005; and Nelson, 2005). These observations have led many
researchers to analyse the statistical properties of inflation persistence over the last two
decades. This statistical issue has special relevance for inflation since in the earlier studies on
developed economies, especially those on the UK and the USA, the unit root status of the
time series were found to be mixed (see Kontonikas, 2004; Bhar and Hamori, 2004; Fountas
et al., 2004; Daal et al., 2005; Conrad and Karanasos, 2006; and Fountas et al., 2006; for
details). Thus, in this context, the span of the data sets used in this study viz.,
January 1970
-
June
2013, is quite large. Hence, structural breaks in the trend function of these
series are likely to occur, and as Perron (1989) first pointed out, disregarding this could
mislead the conclusion on the presence of unit roots in the series.
To test for stationarity of a time series, most often the augmented Dicky-Fuller (ADF; Dicky
and Fuller, 1979) test is used, where the null hypothesis of non-stationarity is tested against
the alternative of stationarity. However, due to the influential work by Perron (1989), the
commonly used ADF test has been criticized because of its bias towards non-rejection of the
null hypothesis of a unit root against the alternative of trend stationarity in the presence of a
structural break in the deterministic trend, as well as for its low power for near integrated
process. Subsequently, Perron (1989, 1990) proposed an alternative unit root test that allows
for the possibility of a structural break in the trend function under both the null and alternative
hypotheses. However, one serious limitation of this test is that it is based on the assumption of
a known break date. Subsequently, Zivot and Andrews (1992), Perron (1997), and Vogelsang
and Perron (1998), among others, have treated the break date to be unknown, which is
endogenously determined from the model.
However, recently Kim and Perron (2009) have pointed out that in all these tests with an
endogenous break point i.e., those by Zivot and Andrews (1992), Perron (1997), and
Vogelsang and Perron (1998), a trend break is not allowed under the null hypothesis. These
tests consider a break in the time series under the alternative only. Hence, tests of this kind are
likely to be affected in terms of size and power. To overcome this limitation, Kim and Perron
(2009) have developed a new test procedure on the line of Perron’s (1989) origina l
formulation of trend break being allowed under both the null and alternative hypotheses, but
the break date is now assumed to be unknown.
Now, prior to applying this unit root test, knowing whether a structural break is present in a
given time series is crucial, especially when the break date is assumed to be unknown, as in
the case with the Kim-Perron (2009) test. Further, in this context, in the absence of a trend
break, the well-known ADF test has the highest power compared to any other alternative test,
and it is the most appropriate test for testing the presence of unit roots in a time series. It is,
therefore, all the more important that the knowledge of the presence or absence of a break in
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
40
the trend function is available before deciding on the appropriate unit root test. However, as
pointed out by Perron and Yabu (2009), testing for a structural break in the trend function
depends on whether the noise component is stationary or non-stationary with unit roots. Thus
there is some sort of a circular problem in testing for these twin issues. To deal with this,
recently Perron and Yabu (2009) proposed a test for structural change in the trend function of
a univariate time series, which can be performed without any prior knowledge on whether the
noise component is stationary or non-stationary containing unit roots8. Since this test is very
general in its approach insofar as the assumption on noise is concerned, we have performed
this test to detect the presence of structural break, if any, in the trend function of a time series.
Thus, in case the Perron-Yabu test suggests that there is no break in the deterministic trend,
we apply the usual ADF test; otherwise, we use the Kim-Perron test to test the null hypothesis
of unit roots against the alternative of stationarity with a break in the deterministic trend
function under both hypotheses9.
Three models - Model I, Model II and Model III - representing a single change in intercept
only, a one-time change in the slope of linear trend function without a change in level, and a
simultaneous change in the intercept and the slope coefficients, respectively, are considered
for the Perron-Yabu test. For our study, we have considered Model III only. The test statistic
values with trimming percentage being 0.15 are reported in Table 3.2.
Country
Perron-Yabu
structural break test
Kim-Perron
unit root test
Estimated break date
Canada 53.029* -21.615* June 1982
France 48.350* -8.000* September 1983
Germany 8.741* -12.948* October 1982
Italy 12.557* -5.034* December 1982 Japan 21.985* -12.279* December 1976
The UK 41.028* -7.767* December 1981
The USA 27.492* -15.488* September 1981
Table 3.2 Results of tests for structural break and unit roots in inflation series.
Notes: *, ** and *** indicate significance at 1%, 5% and 10% levels of significance, respectively.
The Perron-Yabu test statistic values are significant for all seven inflation series, which
clearly establish that either or both of the intercept and slope of the trend function have
undergone structural change. Thus the test confirms the presence of one significant change in
the deterministic trend of inflation for each of the G7 countries.
With this empirical finding on the presence of a structural break in the trend function of
inflation, we have applied the third variant of additive outlier model10
, Model A3 in Kim and
Perron (2009), to each of the seven inflation series to test the null hypothesis of unit roots
with a deterministic trend break against the alternative of stationarity with a break in the
deterministic trend. The values of this test statistic are reported in the 3rd
column of Table 3.2.
The computed values of the t-ratio are compared with the appropriate critical values available
8 Perron and Yabu (2009) have considered a quasi-feasible generalized least squares technique that uses a super-
efficient estimate of the sum of autoregressive parameters, which governs the stationary or integrated behaviour
of a time series. 9 A GAUSS program code for Perron and Yabu (2009) structural break test and a MATLAB code for Kim and
Perron (2009) unit root test in presence of a structural break have been taken from the official website of Pierre
Perron. 10 Kim and Perron (2009) have considered three types of additive outlier models viz., A1, A2 and A3,
representing the occurrence of a structural break in the intercept only, in the slope only, and both in the intercept
and slope coefficients of the trend function, respectively.
-
International Econometric Review (IER)
41
in Perron (1989) and Perron and Vogelsang (1993). The findings clearly suggest that the
inflation series of each of these countries are stationary when accounting for the presence of
one structural break. In each case, the null hypothesis of a unit root with a trend break against
the alternative of a stationary process with a trend break is rejected. We thus conclude that the
underlying data generating process for each of the seven inflation series is a trend stationary
process (TSP), having a structural break in the respective deterministic trend functions.
The stationary time series on inflation for each country, denoted by π t, has then been obtained
after removing the segmented deterministic trend from π t. Specifically, this has been obtained
by regressing π t on an intercept, an intercept dummy, a linear trend term and a trend (linear)
dummy, and then collecting the detrended series, π t, which is now stationary in the sense of
having neither any stochastic trend or any deterministic trend. Thus the obtained stationary
time series on inflation has been used for all subsequent analyses. In this context it may be
noted that the two inflation regimes, defined as π t-1 ≤
0 and π t-1
>
0 in Section 2, essentially
refer to π t-1 being less than or equal to and greater than some positive values, as indicated by
their respective deterministic trend function of each inflation series. Hence the two regimes
are referred to as the low-inflation and high-inflation regimes, respectively.
The estimated break dates are reported in the 4th
column of Table 3.2. The estimated break
dates refer to the period of 1970s and 1980s. Thus these findings provide empirical support to
the general observation that the high phase of inflation during the 1960s-1970s started
reducing substantially towards stabilization in the 1980s. There are quite a few studies
supporting this observation as well. For instance, in their study, Cecchetti and Krause (2001)
have reported that inflation in most developed and developing countries remained at a very
high level during the period of ‘great inflation’ in 1960s, which was coupled with the oil price
shock in 1973. The volatility appears to be more pronounced during that period while
remaining low and stable during the latter half of 1980s due to marked improvement in
macroeconomic performances in those countries. This finding has also been supported by
Stock and Watson (2002), who pointed out that during the period of 1990s, not only volatility
of inflation decreased sharply but also average inflation underwent a downward trend.
Further, Krause (2003) reported that in a cross-section of 63 countries, mean inflation has
fallen from approximately 83 per cent in the pre-1990 period to approximately 9 per cent in
the latter half of 1990s. Given such empirical findings, it is expected that there would be at
least one structural break in the time series on inflation, and this is, in fact, what we have
found11
.
3.2. Empirical Findings on The Impact of Inflation on Inflation Uncertainty
In this section we first present the estimates of the benchmark AR(k)-GARCH(1,1)L(1)
model. Next we discuss the results of estimation and testing of hypotheses of interests
involving the parameters of the proposed model.
3.2.1. The Benchmark AR(k)-GARCH(1,1)L(1) Model
The results of estimation of the benchmark AR(k)-GARCH(1,1)L(1) model specified in
equations (2.1) and (2.2) are reported in Table 3.3. As noted in the preceding section, the
11 Clark (2006) and Levin and Piger (2003) allowed the possibility of structural breaks while examining the
persistence of inflation. Rapach and Wohar (2005) also tested for breaks and found evidence of shift in the level
of inflation in a wide range of European countries.
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
42
optimal lag orders of the autoregressive terms for the seven inflation series have been
obtained by AIC.
Canada France Germany Italy Japan The UK The USA
Conditional mean
ϕ0 0.007 0.006 0.003 -0.004 0.003 -0.012 -0.004
ϕ1 -0.012 0.305 * 0.078 0.266 * 0.082 *** 0.208 * 0.351 *
ϕ2 0.141 * 0.026 0.184 * 0.199 * 0.037 0.298 * 0.016
ϕ3 0.094 ** 0.114 ** 0.110 ** 0.150 * 0.021 0.124 ** -0.008
ϕ4 0.049 0.020 0.058 0.032 0.043 -0.096 ** 0.099 **
ϕ5 0.113 ** -0.018 0.015 -0.016 0.114 ** 0.057 -0.029
ϕ6 0.016 0.052 0.117 * 0.119 ** 0.029 0.114 * 0.019
ϕ7 0.035 0.080 *** 0.038 0.091 *** 0.082 *** - 0.067 ***
ϕ8 - 0.045 0.067 *** 0.032 0.025 - -0.058
ϕ9 - -0.003 - 0.027 0.150 * - 0.062
ϕ10 - 0.057 *** - 0.026 0.044 - 0.055
ϕ11 - - - -0.046 0.051 - -
ϕ12 - - - -0.125 * -0.133 * - -
Conditional variance
ω 0.011 * 0.009 ** 0.013 * 0.001 *** 0.002 *** 0.008 * 0.004 *
α 0.237 * 0.181 * 0.464 * 0.160 * 0.080 ** 0.675 * 0.271 *
β 0.662 * 0.553 * 0.347 * 0.830 * 0.896 * 0.437 * 0.692 *
θ 0.025 ** 0.007 0.023 0.002 -0.007 -0.007 0.024 *
MLLV -75.64 161.00 97.86 182.77 -121.21 -6.780 75.66
Table 3.3 Estimates of the parameters of AR(k)-GARCH(1,1)L(1) model.
Notes: *, ** and *** indicate significance at 1%, 5% and 10% levels of significance, respectively. MLLV is the
maximized log-likelihood value. Some entries in the table are blank since the estimates of the parameters have
been reported, as expectedly, up to the lag (k) values which have been found by the AIC for the mean models of
the respective countries. Evidently, the value of k has not been found to be the same for all countries.
According to the results reported in this table, the autoregressive coefficients are significant in
varying numbers across the seven countries. The usual GARCH parameters viz., ω, α and β
are statistically significant for all inflation series. Additionally, in agreement with the
Friedman-Ball hypothesis, the estimate of the lagged inflation coefficient θ, is positive and
statistically significant only for two countries viz., Canada and the USA, while for the
remaining countries no significant relationship exists. It is quite possible that this finding of
no such significant effect of inflation on its uncertainty for most viz., five, of the G7 countries
may be due to the fact that this model does not factor for regime-specific inflation behaviour,
which might have masked potentially different realizations due to probable regime shift in
inflation series, especially because the span of the data set is quite large
3.2.2. The Proposed DTGARCH(1,1)L(1) Model
The DTGARCH(1,1)L(1) model, as specified in equations (2.3) and (2.4), incorporates
differential behaviour in both inflation and inflation uncertainty from consideration of regime
switching and also allows for the coefficient attached to the lagged inflation term in inflation
uncertainty model specified in equation (2.4) to be different in the two regimes. This model is
locally i.e., regime-wise linear but overall nonlinear in nature. The estimates12
of the
parameters of this model are presented in Table 3.4.
12 For writing the GAUSS codes of DTGARCH (1,1)L(1) model, we have made use of codes from the GAUSS
programs developed by Philip Hans Franses and Dick Van Dijk on different volatility models, which were
downloaded from http://www.few.eur.nl/few/few/people/frances.
http://www.few.eur.nl/few/few/people/frances
-
International Econometric Review (IER)
43
Canada France Germany Italy Japan The UK The USA
Conditional mean for the low-inflation regime
ϕl
0 0.007 0.005 0.010 -0.011 -0.022 -0.053 * -0.001
ϕl
1 0.069 0.326 * 0.141 0.261 * -0.020 -0.050 0.394 *
ϕl
2 0.087 -0.033 0.134 ** 0.283 * 0.014 0.328 * 0.025
ϕl
3 0.153 ** 0.029 0.163 ** 0.046 0.022 0.215 * -0.079
ϕl
4 0.078 0.171 ** 0.109 *** -0.026 0.013 -0.084 0.155 **
ϕl
5 0.160 ** -0.092 -0.011 0.018 0.109 *** -0.055 -0.031
ϕl
6 -0.073 -0.031 0.120 ** 0.113 ** -0.009 0.178 * 0.121 ***
ϕl
7 0.000 0.116 0.052 0.186 * 0.100 *** - -0.021
ϕl
8 - 0.057 0.089 *** -0.035 0.009 - -0.106
ϕl
9 - 0.041 - -0.043 0.112 ** - 0.130 **
ϕl
10 - 0.075 - 0.049 0.094 - 0.086
ϕl
11 - - - -0.029 0.013 - -
ϕl
12 - - - -0.152 * -0.084 - -
Conditional mean for the high-inflation regime
ϕh
0 0.018 -0.015 0.014 -0.006 -0.003 -0.018 -0.006
ϕh
1 -0.044 0.401 * 0.034 0.221 ** 0.121 0.425 * 0.325 *
ϕh
2 0.276 * 0.141 *** 0.208 * 0.217 * 0.094 0.263 * 0.020
ϕh
3 0.032 0.169 ** 0.047 0.043 0.097 0.029 0.019
ϕh
4 -0.004 -0.072 -0.003 0.157 * 0.099 -0.031 0.040
ϕh
5 0.055 0.023 0.030 0.089 *** 0.129 *** 0.147 * -0.013
ϕh
6 0.020 0.125 0.114 ** 0.256 * 0.001 0.138 * -0.012
ϕh
7 0.089 0.021 0.005 -0.020 0.076 - 0.146 **
ϕh
8 - 0.042 0.072 -0.034 -0.030 - -0.001
ϕh
9 - -0.047 - 0.137 * 0.144 ** - -0.040
ϕh
10 - -0.007 - 0.139 ** 0.062 - 0.027
ϕh
11 - - - -0.157 * 0.060 - -
ϕh
12 - - - -0.251 * -0.195 * - -
Conditional variance for the low-inflation regime
ωl 0.027
** 0.004 ** 0.011 *** 0.002 0.006 0.002 0.004 ***
αl 0.605
* 0.363 ** 0.604 * 0.255 *** 0.000 0.640 * 0.304 *
βl 0.468
* 0.710 * 0.492 * 0.323 * 0.555 * 0.529 * 0.619 *
θl 0.111
** 0.034 *** 0.046 -0.055 * -0.154 ** -0.017 0.010
Conditional variance for the high-inflation regime
ωh 0.002 0.033
* 0.008 0.000 0.000 0.009 ** 0.000
αh 0.000 0.120 0.000 0.948
* 0.222 ** 0.513 * 0.001
βh 0.658
* 0.001 0.280
* 0.020 0.837
* 0.000 0.655
*
θh 0.119
* -0.039 0.154 * 0.153 * -0.003 0.198 * 0.124 *
MLLV -63.51 172.46 104.54 198.89 -116.26 16.39 88.40
Wald test for
H0: θl = θh 0.02 2.69 5.42 ** 33.89 * 3.51 *** 13.40 * 11.19 *
Table 3.4 Estimates of the parameters of the DTGARCH(1,1)L(1) model.
Notes: *, ** and *** indicate significance at 1%, 5% and 10% levels of significance, respectively. MLLV is the
maximized log-likelihood value. Some entries in the table are blank since the estimates of the parameters have
been reported, as expectedly, up to the lag (k) values which have been found by the AIC for the mean models of
the respective countries. Evidently, the value of k has not been found to be the same for all countries.
Several findings from the estimation results are worth mentioning: First, the estimates of the
parameters in conditional mean clearly establish regime switching behaviour in inflation for
all seven series since some of the own lags in both the regimes are found to be statistically
significant. Second, looking at the parameters of the model for the conditional variance in the
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
44
two inflation regimes, we first note that in case of Canada, France, Germany, Japan, and the
USA, only one of the parameters of αl and α
h is significant, while for Italy and the UK, both
parameters are significant. Further, only one of the parameters βl and β
h are significant for the
three countries viz., France, Italy, and the UK, and for the remaining ones i.e., Canada,
Germany, Japan, and the USA, both coefficients are significant. Considering two regimes
based on a past level of inflation for the conditional variance which measures inflation
uncertainty, is found to be relevant and statistically meaningful. Regarding the effect of
inflation on inflation uncertainty, i.e., θl and θ
h, we note that both these coefficients are
significant in two countries only, namely, Canada and Italy, while for the remaining five
countries, at least one of θl and θ
h is significant. This establishes the fact that the effect of
inflation on inflation uncertainty is also regime specific. Hence, on the whole, any policy
measure on inflation should be based, inter alia, on regime consideration of inflation.
Finally, the most important hypothesis for the proposed model is whether inflation affects
inflation uncertainty differently in the two inflation regimes. In terms of the parameters, the
null and alternative hypotheses are H0: θ
l =
θ
h and H1:
θ
l ≠
θ
h, respectively. This null
hypothesis has been tested by using the Wald test, and the test statistic values for the seven
series are reported in the last row of Table 3.4. The results of the Wald test show that the null
hypothesis is rejected for all but Canada and France. Regarding the latter two countries, we
can conclude that in the case of Canada, inflation has a positive impact on inflation
uncertainty and does not vary with the regime change while for France, the only conclusion
that can be drawn is that the coefficients do not change significantly between the two regimes.
The Wald test results thus suggest that significant difference in the two regimes exists insofar
the effect of inflation on inflation uncertainty is concerned in case of five members of the G7
countries. These countries are Germany, Italy, Japan, the UK, and the USA.
The positive and significant values of θh in four of these countries viz., Germany, Italy, the
UK, and the USA indicate that inflation increases inflation uncertainty at the high-inflation
regime in each of these countries, while the effect is insignificant for Germany, the UK, and
the USA at the low-inflation regime and negative for Italy. This evidence for Germany, the
UK, and the USA thus supports the findings of Ungar and Zilberfarb (1993) viz., that inflation
affects inflation uncertainty only in the high-inflation regime but not in the low-regime. The
finding that θˆl is negative and significant for Italy and Japan is, however, somewhat unusual
although not exceptional. This means that at the low-regime inflation has a negative effect on
inflation uncertainty or in other words, a reduction in inflation in the low-regime increases
inflation uncertainty, and thus the Friedman-Ball hypothesis fails to hold for these two
countries. Such evidence has been found by a few other studies as well – although with
different volatility specifications. For instance, based on a study on 12 European Monetary
Union countries, Caporale and Kontonikas (2009) have found that during the post-1999
period when average inflation was low, a further reduction in inflation led to an increase
rather than a decrease in inflation uncertainty for a number of countries in the Euro Zone. In a
recent study based on monthly data on US inflation over the period from 1926 to 1992,
Hwang (2001) has found that inflation affects its uncertainty weakly and negatively during
the periods of both high and low inflation.
We report on the Ljung-Box test statistic values based on residuals of this model for all G7
countries in Table 3.5.
The test statistic has been computed for both the standardized and squared standardized
residuals. It is evident from Table 3.5 that none of these are significant, and hence we
-
International Econometric Review (IER)
45
conclude that the proposed models for both the conditional mean and conditional variance of
inflation along with the chosen lag values are adequate for all G7 countries.
Country Q(1) Q(5) Q(10) Q2(1) Q
2(5) Q
2(10)
Canada 0.007 3.746 7.027 0.670 1.386 4.238
France 0.107 1.975 4.710 0.014 4.348 5.481
Germany 0.369 2.123 4.823 0.638 3.656 7.972
Italy 0.024 2.434 11.420 2.281 5.418 12.963
Japan 0.027 0372 1.480 0.002 2.112 8.114
The UK 0.001 2.676 6.196 0.003 2.303 11.887
The USA 0.014 2.604 5.050 1.501 7.512 9.926
Table 3.5 Results of the Ljung-Box test with standardized residuals and squared standardized residuals.
Notes: *, ** and *** indicate significance at 1%, 5% and 10% levels of significance, respectively. Q(.) and Q2(.)
denote the Ljung-Box test for autocorrelation in standardized residuals and squared standardized residuals,
respectively.
Finally, we make a comparison between the benchmark AR(k)-GARCH(1,1)L(1) model and
the proposed DTGARCH(1,1)L(1) model by the likelihood ratio (LR) test to find if
introduction of regimes based on low and high inflation in both conditional mean and
conditional variance has led to any significant gain in understanding the effects of inflation on
inflation uncertainty. We report the LR test statistic values in Table 3.6.
Country
AR(k)-GARCH(1,1)L(1)
versus
DTGARCH(1,1)L(1)
Canada 24.26 **
France 22.92 ***
Germany 13.36
Italy 32.24 **
Japan 9.90
The U.K. 46.34 * The U.S.A. 25.48 **
Table 3.6 LR test statistic values.
Notes: *, ** and *** indicate significance at 1%, 5% and 10% levels of significance, respectively. Q(.) and Q2(.) denote the Ljung-Box test for autocorrelation in standardized residuals and squared standardized residuals,
respectively.
We observe from this table that the LR test statistic values are significant for five of the G7
countries viz., Canada, France, Italy, the UK, and the USA, and hence we can conclude that
the proposed model explains the relationship ‘better’ than the benchmark model.
4. CONCLUSIONS
The effect of inflation on inflation uncertainty has been studied for the G7 countries in terms
of a model called the DTGARCH(1,1)L(1) model, where the conditional mean as well as the
conditional variance of inflation are based on a consideration of two regimes for inflation, and
the conditional variance specification for each regime is assumed to be GARCH with an
additional term of inflation with a lag of 1. The regimes are determined by the level of
(stationary) inflation of the preceding lag being negative or positive.
The findings on the twin issues of structural break and stationarity of the series, based on the
recently-developed tests of Perron and Yabu (2009) and Kim and Perron (2009), are that all of
the series are trend stationary with a break in their respective trend functions. The estimated
break date thus obtained for each country seems to broadly give empirical support to the
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
46
phenomenon of ‘great inflation’, which refers to the high and volatile inflation that occurred
in the mid-1960s and lasted for almost twenty years
Regarding the nature of the relationship, the empirical findings clearly show that the impact
of inflation on inflation uncertainty is different in the two regimes for five of the G7 countries
viz., Germany, Italy, Japan, the UK, and the USA. For Canada, the Friedman-Ball hypothesis
for a positive effect of inflation on its uncertainty holds, but this effect is invariant to the two
regimes. Further, the Friedman-Ball hypothesis holds only in the high-inflation regime for
Germany, Italy, the UK, and the USA. On the other hand, the relationship is found to be
insignificant for Germany, the UK, and the USA in the low-inflation regime, while it is
negative and significant for Italy and Japan. This suggests that in the low-inflation regime a
decline in inflation has led to an increase in inflation uncertainty for Italy and Japan, and this
obviously counters the Friedman-Ball hypothesis. Thus, these findings, on one hand, provide
strong empirical support to the Friedman-Ball hypothesis for high-inflation regimes of most
of the G7 countries, and on the other, clearly establish the importance of considering regime-
specific behaviour in modelling the impact of inflation on inflation uncertainty.
REFERENCES
Arghyrou, M., C. Martin and C. Milas (2005). Non-linear inflationary dynamics: evidence
from the UK. Oxford Economic Papers, 57 (1), 51-69.
Baillie, R., C.F. Chung and M. Tieslau (1996). Analyzing inflation by the fractionally
integrated ARFIMA-GARCH model. Journal of Applied Econometrics, 11 (1) , 23-40.
Ball, L. (1992). Why does higher inflation raise inflation uncertainty? Journal of Monetary
Economics, 29 (3), 371-378.
Berndt, E.K., B.H. Hall, R.E. Hall and J.A. Hausman (1974). Estimation and inference in
nonlinear structural models. Annals of Economic and Social Measurement, 3(4), 103-
116.
Bhar, R. and S. Hamori (2004). The link between inflation and inflation uncertainty: evidence
from G7 countries. Empirical Economics, 29 (4) , 825-853.
Blinder, A.S. (1982). The anatomy of double-digit inflation in the 1970s. In NBER Chapters,
in: Inflation: Causes and Effects, 261-382. National Bureau of Economic Research.
Bollerslev, T. (1986). Generalised autoregressive conditional heteroskedasticity. Journal of
Econometrics, 31 (3), 307-327.
Bredin, D. and S. Fountas (2005). Macroeconomic uncertainty and macroeconomic
performance: Are they related? The Manchester School, 73 (s1), 58-76.
Brooks, C. (2001). A double-threshold GARCH model for the French/Deutschmark exchange
rate. Journal of Forecasting, 20 (2), 135-143.
Caglayan, M., and A. Filiztekin (2003). Nonlinear impact of inflation on relative price
variability. Economics Letters, 79 (2), 213-218.
-
International Econometric Review (IER)
47
Caporale, G.M. and A. Kontonikas (2009). The Euro and inflation uncertainty in the
European Monetary Union. Journal of International Money and Finance, 28 (6), 954-
971.
Cecchetti, S. G. and S. Krause (2001). Financial structure, macroeconomic stability and
monetary policy. NBER Working Paper 8354.
Chan, K. S. and H. Tong (1986). On estimating thresholds in autoregressive models. Journal
of time series analysis, 7 (3), 179-190.
Chang, K.L. (2012). The impacts of regime-switching structures and fat-tailed characteristics
on the relationship between inflation and inflation uncertainty. Journal of
Macroeconomics, 34 (2), 523-536.
Chang, K.L. and C.W. He (2010). Does the magnitude of the effect of inflation uncertainty on
output growth depend on the level of inflation? Manchester School, 78 (2), 126-148.
Chen, C.W. (1998). A Bayesian analysis of generalized threshold autoregressive models.
Statistics & probability letters, 40 (1), 15-22.
Chen, C.W. and M.K. So (2006). On a threshold heteroscedastic model. International Journal
of Forecasting, 22 (1), 73-89.
Chen, C.W., T.C. Chiang and M.K. So (2003). Asymmetrical reaction to US stock-return
news: evidence from major stock markets based on a double-threshold model. Journal
of Economics and Business, 55 (5), 487-502.
Chen, C.W., M.J. Yang, R. Gerlach and H. Jim Lo (2006). The asymmetric ractions of mean
and volatility of stock returns to domestic and international information based on a four-
regime double-threshold GARCH model. Physica A: Statistical Mechanics and its
Applications, 366 , 401-418.
Chen, S.W., C.H. Shen and Z. Xie (2008). Evidence of a nonlinear relationship between
inflation and inflation uncertainty: The case of the four little dragons. Journal of Policy
Modeling, 30 (2) , 363-376.
Clarida, R., J. Gali and M. Gertler (2000). Monetary policy rules and macroeconomic
stability: evidence and some theory. Quarterly Journal of Economics, 115 (1), 147-180.
Clark, T.E. (2006). Disaggregate evidence on the persistence of consumer price inflation.
Journal of Applied Econometrics, 21(5), 563-587.
Conrad, C. and M. Karanasos (2006). Dual long memory in inflation dynamics across
countries of the Euro area and the link between inflation uncertainty and
macroeconomic performance. Studies in Nonlinear Dynamics and Econometrics, 9 (4),
1-38.
Cosimano, T.F. and D.W. Jansen (1988). Estimates of the variance of U.S. inflation based
upon the ARCH model. Journal of Money, Credit and Banking, 20 (3), 409-421.
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
48
Daal, E., A. Naka and B. Sanchez (2005). Re-examining inflation and inflation uncertainty in
developed and emerging countries. Economics Letters, 89 (2), 180-186.
DeLong, J.B. (1997). America's peacement inflation. In Reducing Inflation: Motivation and
Strategy, ed. C.D. Romer and D.H. Romer, University of Chicago Press, 247-280.
Dickey, D.A. and W.A. Fuller (1979). Distribution of the estimators for autoregressive time
series with a unit root. Journal of the American Statistical Association, 74, 427-431.
Engle, R.F. (1982). Autoregressive conditional heteroskedasticity with estimates of the
variance of UK inflation. Econometrica, 50 (4), 987-1007.
Engle, R.F. (1983). Estimates of the Variance of US Inflation Based upon the ARCH Model.
Journal of Money, Credit and Banking, 15 (3), 286-301.
Evans, M. and P. Wachtel (1993). Were price changes during the Great Depression
anticipated? Evidence from nominal interest rates. Journal of Monetary Economics, 32
(1), 3-34.
Fischer, S. (1993). The role of macroeconomic factors in growth. Journal of monetary
economics, 32 (3), 485-512.
Fountas, S. (2001). The relationship between inflation and inflation uncertainty in the UK:
1885–1998. Economics Letters, 74 (1), 77-83.
Fountas, S., A. Ioannidis and M. Karanasos (2004). Inflation, inflation uncertainty and a
common European monetary policy. Manchester School, 72 (2), 221-242.
Fountas, S. and M. Karanasos (2007). Inflation, output growth, and nominal and real
uncertainty: empirical evidence for the G7. Journal of International Money and
Finance, 26 (2), 229-250.
Fountas, S., M. Karanasos and M. Karanassou (2000). A GARCH model of inflation and
inflation uncertainty with simultaneous feedback. Department of Economics, University
of York , Discussion Papers 00/24.
Fountas, S., M. Karanasos and J. Kim (2002). Inflation and output growth uncertainty and
their relationship with inflation and output growth. Economics Letters, 75 (3), 293-301.
Fountas, S., M. Karanasos and J. Kim (2006). Inflation uncertainty, output growth uncertainty
and macroeconomic performance. Oxford Bulletin of Economics and Statistics, 68 (3),
319-343.
Friedman, M. (1977). Nobel lecture: inflation and unemployment. Journal of Political
Economy, 85 (3), 451-472.
Granger, C.W. and T. Teräsvirta (1993). Modelling non-linear economic relationships. OUP
Catalogue.
-
International Econometric Review (IER)
49
Grier, K.B. and M.J. Perry (1998). On inflation and inflation uncertainty in the G7 countries.
Journal of International Money and Finance, 17 (4) , 671-689.
Hamilton, J. (2001). A parametric approach to flexible nonlinear inference. Econometrica, 69
(3), 537-573.
Hwang, Y. (2001). Relationship between inflation rate and inflation uncertainty. Economics
Letters, 73 (2), 179-186.
Kim, C.J. (1993). Unobserved-component models with Markov switching heteroskedasticity:
changes in regime and link between inflation rates and inflation uncertainty. Journal of
Business and Economic Statistics, 11 (3), 341-349.
Kim, C.J. and C.R. Nelson (1999). Has the US economy become more stable? A Bayesian
approach based on a Markov-switching model of the business cycle. Review of
Economics and Statistics, 81 (4), 608-616.
Kim, D. and P. Perron (2009). Unit root tests allowing for a break in the trend function at an
unknown time under both the null and alternative hypothesis. Journal of Econometrics,
148 (1), 1-13.
Kontonikas, A. (2004). Inflation and inflation uncertainty in the United Kingdom, evidence
from GARCH modelling. Economic Modelling, 21 (3), 525-543.
Krause, S. (2003). Measuring monetary policy efficiency in European Union countries.
Department of Economics, Emory University, Working Paper No. 03-11.
Lanne, M. (2006). Nonlinear dynamics of interest rate and inflation. Journal of Applied
Econometrics, 21 (8), 1157-1168.
Levin, A. T. and J. Piger (2003). Is inflation persistence intrinsic in industrial economies?
Federal Reserve Bank of St. Louis Working Paper 2002-023E.
Li, C.W. and W.K. Li (1996). On a double-threshold autoregressive heteroscedastic time
series model. Journal of Applied Econometrics, 11 (3), 253-274.
Meltzer, A.H. (2005). Origins of the great inflation. Federal Reserve Bank of St. Louis
Review, 87 (2), 145-175.
Nelson, E. (2005). The great inflation of the seventies: what really happened? The B.E.
Journal of Macroeconomics, 5 (1), 1-50.
Orphanides, A. (2003). The quest for prosperity without inflation. Journal of Monetary
Economics, 50 (3), 633-663.
Perron, P. (1997). Further evidence on breaking trend functions in macroeconomic variables.
Journal of Econometrics, 80 (2) , 355-385.
Perron, P. (1990). Testing for a unit root in a time series with changing mean. Journal of
Business & Economic Statistics, 8 (2), 153-162.
-
Chowdhury and Sarkar-Eff. of Infl. on Infl. Uncer. in G7 Count.: A Double Threshold GARCH Model
50
Perron, P. (1989). The great crash, the oil price shock, and the unit root hypothesis.
Econometrica, 57 (6), 1361-1401.
Perron, P. and T.J. Vogelsang (1993). The great crash, the oil price shock and the unit root
hypothesis: Erratum. Econometrica, 61 (1), 248-249.
Perron, P. and T. Yabu (2009). Testing for shifts in trend with an integrated or stationary
noise component. Journal of Business and Economic Statistics, 27 (3), 369-396.
Rapach, D. and M. Wohar (2005). Regime changes in international real interest rates: Are
they a monetary phenomenon? Journal of Money, Credit and Banking, 37(5), 887-906.
Stock, J. H., and Watson, M. W. (1996). Evidence on structural instability in macroeconomic
time series relations. Journal of Business & Economic Statistics, 14(1) , 11-30.
Stock, J.H. and M.W. Watson (2002). Has the business cycle changed and why? NBER
Macroeconomics Annual 2002, 17, 159-230.
Teräsvirtra, T. (1998). Modelling economic relationships with smooth transition regressions.
In Handbook of Applied Economic Statistics, ed. A. Ullah and D.E. Giles, 507-552. New
York: Marcel Dekker.
Tong, H. and K.S. Lim (1980). Threshold autoregression, limit cycles and cyclical data.
Journal of the Royal Statistical Society, Series B (Methodological), 42 (3), 245-292.
Ungar, M. and B. Zilberfarb (1993). Inflation and its unpredictability - theory and empirical
evidence. Journal of Money, Credit and Banking, 25 (4), 709-720.
Vogelsang, T.J. and P. Perron (1998). Additional tests for a unit root allowing for a break in
the trend function at an unknown time. International Economic Review, 39 (4), 1073-
1100.
Wilson, B.K. (2006). The links between inflation, inflation uncertainty and output growth:
New time series evidence from Japan. Journal of Macroeconomics, 28 (3), 609-620.
Yang, Y.L. and C.L. Chang (2008). A double-threshold GARCH model of stock market and
currency shocks on stock returns. Mathematics and Computers in Simulation, 79(3),
458-474.
Zivot, E. and D.K. Andrews (1992). Further evidence on the great crash, the oil price shock
and the unit root hypothesis. Journal of Business and Economic Statistics, 10 (3) , 251-
270.